Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
De la co-construction à la co-évolution entre acteurs humains et IA au sein du cycle de personnalisation des EIAH
No full textThis manuscript presents my main contributions in the field of Technology-Enhanced Learning (TEL), and more specifically in the use of Artificial Intelligence (AI) for personalized learning, closely aligned with learners’ needs and teachers’ constraints. Within this context, I address two major themes.First, I propose a Knowledge Engineering-based approach to support teachers in the co-construction of personalized learning environments. This includes the development of meta-models for modeling pedagogical knowledge and designing authoring tools, as well as the integration of these models into adapted environ-ments. The aim is not to create automated systems that replace teachers, but rather adaptable tools aligned with their practices, enabling the generation and recommendation of learning activities tailored to learner profiles and specific pedagogical constraints. In recent years, these models and tools have been furtherdeveloped to support the implementation of Competency-Based Education and to fully leverage its potential within the personalization cycle of TEL.Second, I explore approaches based on the collection and analysis of learning traces, drawing on Knowledge Engineering, as well as Data Mining and Machine Learning techniques. This has led to the development of platforms that, on the one hand, allow learning data analysis without requiring technical expertise, and, on the other, facilitate the capitalization and sharing of such analyses.This work has also resulted in the design of recommender systems leveraging competency models, integrating both top-down approaches (expert knowledge) and bottom-up approaches (data-driven discovery). Furthermore, analyzing the operational traces of AI engines has enabled the proposal of initial mechanismsfor explainable AI in these systems.Building on these contributions, developed within national and international projects, a research agenda is also presented. It focuses on the following directions : the design of rich and adapted explainable AI mechanisms within the personalization cycle, the development of hybrid AI to detect and leverage learners’ sense of competence, the co-evolution of the Human-AI relationship within TEL, and the enrichment of our models through the paradigm of active learning in AI. These perspectives aim to better integrate AI capabilities into TEL while ensuring their alignment with the pedagogical and human realities of educational stakeholders.Ce manuscrit présente nos principales contributions dans le domaine des Environnements Informatiques pour l’Apprentissage Humain (EIAH), et plus particulièrement dans l’exploitation de l'Intelligence Artificielle (IA) pour la personnalisation de l’apprentissage, en lien étroit avec les besoins des apprenants et les contraintes des enseignants. Dans ce cadre, nous avons abordé deux grandes thématiques.Premièrement, nous avons proposé une approche fondée sur l’Ingénierie des Connaissances pour accompagner les enseignants dans la co-construction d'environnements personnalisés d’apprentissage. Cela comprend l’élaboration de méta-modèles exploités dans la modélisation des connaissances pédagogiques et dans la conception d’outils auteurs, ainsi que l’intégration de ces modèles dans des environnements adaptés. L’objectif est de proposer non pas des systèmes automatiques remplaçant l’enseignant, mais des outils adaptables à leurs pratiques, permettant de générer et recommander des activités pédagogiques en fonction des profils apprenants et des contraintes pédagogiques spécifiques. Ces modèles et outils ont été enrichis ces dernières années pour favoriser la mise en œuvre de l'Approche par Compétences dans l'enseignement et en exploiter toute la richesse lors du cycle de personnalisation des EIAH.Deuxièmement, nous avons exploré des approches fondées sur la collecte et l’analyse des traces d’apprentissage, en nous appuyant de même sur l'ingénierie des connaissances, mais également sur des techniques de fouille de données et d'apprentissage machine. Cela a permis de proposer des plateformes permettant, d'une part, l'analyse des données d'apprentissage sans expertise technique, et d'autre part, la capitalisation et le partage de ces analyses. Ces travaux ont également mené à la construction de systèmes de recommandations exploitant des modèles de compétences, en intégrant à la fois des approches top-down (connaissances expertes) et bottom-up (découverte à partir des données). Enfin, l'exploitation des traces de fonctionnement de nos moteurs d'IA nous a permis de proposer des premiers mécanismes d'IA explicables pour ces moteurs.À partir de ces contributions développées dans le cadre de projets nationaux et internationaux, un plan de recherche est également présenté. Il s’articule autour des axes suivants : la proposition de mécanismes d'IA explicables riches et adaptés au sein du cycle de personnalisation, la construction d'une IA hybride pour détecter et exploiter le sentiment de compétences chez les apprenants, la co-évolution du couple Humain-IA au sein des EIAH, et enfin l'enrichissement de nos modèles via le paradigme d'IA d'apprentissage actif. Ces perspectives visent à mieux intégrer les capacités de l’IA dans les EIAH tout en garantissant leur adéquation aux réalités pédagogiques et humaines des acteurs de l’éducation
Sobolev estimates for the Keller-Segel system and applications to the JKO scheme
No full textWe prove L^{\infty}_{t} W^{1,p} Sobolev estimates in the Keller-Segel system by proving a functional inequality, inspired by the Brezis-Gallouët-Wainger inequality. These estimates are also valid at the discrete level in the Jordan-Kinderlehrer-Otto (JKO) scheme. By coupling this result with the diffusion properties of a functional according to Bakry-Emery theory, we deduce the L^{2}_{t} H^{2}_{x} convergence of the scheme, thereby extending the recent result of Santambrogio and Toshpulatov in the context of the Fokker-Planck equation to the Keller-Segel system
Generalized Non-Hermitian Hamiltonian for Guided Resonances in Photonic Crystal Slabs
No full textWe develop a generalized non-Hermitian Hamiltonian formalism for guided resonances in photonic crystal slabs, derived directly from Maxwell's equations through a systematic guided-mode expansion. By expanding the electromagnetic fields over the complete mode basis of an unpatterned slab and systematically integrating out radiative Fabry--Pérot channels, we obtain the analytical operator structure of the Hamiltonian, which treats guided-mode coupling and radiation losses on equal footing. The resulting Hamiltonian provides explicit expressions for both dispersive and radiative coupling terms in terms of modal overlap integrals and Fourier components of the permittivity modulation. For specific geometries, the Hamiltonian coefficients can be extracted from full-wave simulations enabling accurate modeling without phenomenological assumptions. As a case study, we investigate hexagonal lattices with both preserved and broken symmetry, demonstrating predictive agreement for complex band structures, near-field distributions, and far-field polarization patterns. In particular, the formalism reproduces symmetry-protected bound states in the continuum (BICs) at the point, accidental off- BICs near the point, and the emergence of chiral exceptional points (EPs). It also captures the tunable behavior of eigenmodes near the point, including Dirac-point shifts and the emergence of quasi-BICs or bandgap openings, depending on the nature of symmetry breaking. We further demonstrate in the Appendix that the same formalism extends naturally to other symmetry classes, including (1D grating) and (square lattice) photonic crystal slabs. This approach enables predictive and efficient modeling of complex photonic resonances, revealing their topological and symmetry-protected characteristics in non-Hermitian systems
Token positional games
No full textThe classical Maker-Breaker positional game is played on a board which is a hypergraph H, with two players, Maker and Breaker, alternately claiming vertices of H until all the vertices are claimed. When the game ends, Maker wins if she has claimed all the vertices of some edge of H; otherwise, Breaker wins. Playing this game in real life can be done by placing tokens on the vertices of the board. In this paper, we study the unfortunate case in which one or both players do not have enough tokens to cover all the vertices and, as such, will have to move their tokens around at some point instead of placing new ones. There may be a bias, in that Maker and Breaker do not necessarily have the same amount of tokens. The present paper initiates the study of this generalization of positional games, called token positional games.A particularly interesting case is when Maker has a winning strategy in the classical game: what is the lowest number of tokens with which she still wins against Breaker's unlimited stock? We notably show that, for k-uniform hypergraphs on an arbitrarily large number n of vertices, this number equals k if k ∈ {2, 3} but can vary from k to Ω(n) if k ≥ 4. From an algorithmic point of view, PSPACE-hardness in general is inherited from classical positional games, but we get a polynomial-time algorithm to solve the case where Breaker only has one token. We also establish EXPTIME-completeness for a "token sliding" variation of the game.</div
Filtrations and asymptotic geometry of non-Archimedean norms on section rings
No full textThis article is concerned with the metric study of a construction of Gérardin of the action of the boundary at infinity of the space of norms on a non-Archimedean vector space, and its generalisation to graded algebras. Namely, given (X,L) a polarised variety over an arbitrary non-Archimedean field, we show that there is a jointly d_1-contracting action of the space of filtrations of the section ring R(X,L) on the space of graded norms on R(X,L). This naturally yields non-Archimedean geodesic rays and infinite-dimensional flats in this setting, generalising previous work of the author and Witt Nyström. It is further shown that relative limit measures converge along geodesic rays, providing a result on the d_p-radial geometry of graded norms, analogous to a recent result of Finski in the Archimedean case
Renaming in distributed certification
No full textInternational audienceLocal certification is the area of distributed network computing asking the following question: How to certify to the nodes of a network that a global property holds, if they are limited to a local verification? In this area, it is often essential to have identifiers, that is, unique integers assigned to the nodes. In this short paper, we show how to reduce the range of the identifiers, in three different settings. More precisely, we show how to rename identifiers in the classical local certification setting, when we can (resp. cannot) choose the new identifiers, and we show how a global certificate can help to encode very compactly a new identifier assignment that is not injective in general, but still useful. We conclude with a number of applications of these three results
Acceleration of implicit schemes for large systems of delay differential equations
No full textInternational audienceThe objective is to accelerate numerical implicit schemes for solving large linear or nonlinear delay differential equations. These schemes require solving large linear or nonlinear systems at each integration step, making effective initial guesses critical for rapid convergence. For nonlinear problems, an inexact Newton method is used, whose efficiency depends heavily on the quality of these initial guesses. To generate them, line search or trust-region algorithms are employed -each involving the solution of large linear systems. These linear systems are solved using a Krylov subspace method. Initial guesses are constructed via a Petrov-Galerkin process applied to low-dimensional approximation subspaces derived from previous steps. Error estimates are provided, linking the accuracy of the initial guesses to the timestep size, the scheme's order, and the subspace dimension. Numerical experiments show speedups of up to two orders of magnitude over standard predictor-based methods, when those converge
Stiffness-sensitive gene regulation in human mesenchymal stem cells: Modelling mechanotransduction to predict mineralization and bone protein expression
No full textInternational audienceThe goal of our study was to establish how a specific part of the bone Gene Regulatory Network (GRN) controls mineralization in response to stiffness. We hypothesized that a system of differential equations model stiffness-sensitive gene regulation in human mesenchymal stem cells through the epistatic genetic interactions between stiffness (e.g. WNT-β catenin pathway) and five of the main transcription factors and bone proteins (e.g. RUNX2, BSP, OSX, OC, and OPN). To test this hypothesis, we (i) performed in-vitro experiments culturing bone cells on different stiffness, (i) adapted our previously published model from being continuously time-dependent to continuously stiffness-sensitive, and (iii) simulated protein production in function of stiffness and other protein production from the best estimate of parameters coming from the experimental work. Our experimental findings reveal a non-parametric relationship between stiffness and RUNX2 production, with no discernible linear trends for other proteins. Modeling results demonstrate that continuous variations in stiffness enable simulation of bone GRN gene expression, fitting our novel experimental dataset. Specifically, our computational results indicate that OPN production peaks at low stiffness (8 kPa), while RUNX2, OSX, and OC achieve maximum production at higher stiffness levels (64 kPa). This alignment underscores the model's capacity to replicate experimental data accurately. Additionally, our approach predicts that WNT-β-catenin activation serves as an enhancer for OPN and BSP production. The model also highlights a negative feedback-like interaction between OC and BSP production. Stiffness variations were shown to have a significant impact on OC and BSP production and a moderate effect on OPN production. By employing a stiffness-sensitive gene regulation model, we provide insights into one of the mineralization patterns through the prediction of bone protein expression dynamics
(Anti)Ferroelectric HfZrO2: from non volatile memory to energy storage applications
No full textInternational audienceFluorite-structured HfZrO2 and ZrO2 thin films are promising for both non-volatile memory and energy storage applications. Antiferroelectric (AFE) capacitors were fabricated by PE-ALD and annealed using BEOL-compatible thermal budgets. All devices exhibit AFE doublehysteresis loops, with polarization strongly dependent on film thickness and annealing temperature. Phase-field modeling reveals thicknessdriven phase transitions enabling low-voltage, high-polarization operation.These characteristics are relevant for future AFE FET. Energy storage performances of linear dielectric (LD), ferroelectric (FE), and AFE films are compared. While FE films show high energy density but large losses, LDs offer high efficiency with low storage density. AFEZrO2 provides an optimal balance, reaching up to 84 J/cm3 with 75% efficiency. These results position HfZrO2 as a multifunctional platform bridging memory and energy storage technologies